# Open Catalyst 2020 (OC20) S2EF 20M Training Set, ColabFit Conversion

## Facts

- Provider: Meta FAIR and Carnegie Mellon University (Chanussot et al.); ColabFit conversion
- License: CC-BY-4.0 (https://raw.githubusercontent.com/facebookresearch/fairchem/main/docs/catalysts/datasets/oc20.md)
- Size: 203 files
- Format: Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/colabfit/OC20_S2EF_train_20M
- Access: public, readable without an account
- DOI: https://doi.org/10.60732/9f03e9be
- Category: Chemistry & Materials
- Source: https://doi.org/10.1021/acscatal.0c04525
- Page: https://scigantic.com/archive/1166ee02-b599-5977-9d96-8eab8a3a7c08
- Keywords: Chemistry & Materials, DFT, interatomic potentials, Parquet, Hugging Face

## About

The 20 million structure training split of the Open Catalyst 2020 (OC20) Structure to Energy and Forces (S2EF) task, released by Meta FAIR and Carnegie Mellon University for machine learning interatomic potentials in heterogeneous catalysis. Each configuration is an adsorbate on an inorganic catalyst surface slab, labelled with DFT (RPBE, VASP) total energy, adsorption energy and per-atom forces, plus OC20 mapping metadata such as adsorbate id, Materials Project bulk id, Miller index and shift. ColabFit's card counts 20,000,000 configurations and about 1.47 billion atoms over 53 elements; the repository uses about 85.6 GB (measured 2026-11-14). It is the standard large training set for S2EF models such as GemNet, EquiformerV2 and eSCN. This is the ColabFit Exchange conversion: configurations are Parquet files under co/ (structure, positions, cell, computed properties and metadata per row) with dataset-level information in ds.parquet; read them with pyarrow, pandas, polars or Hugging Face datasets, following the ColabFit configuration schema at materials.colabfit.org/docs. Upstream OC20 is released by Meta under CC-BY-4.0; cite Chanussot et al., ACS Catalysis 2021.

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("colabfit/OC20_S2EF_train_20M")
print(ds)
```

## Data structure

File types in sampled files: Parquet

Example file: `co/co_0.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/1166ee02-b599-5977-9d96-8eab8a3a7c08
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/1166ee02-b599-5977-9d96-8eab8a3a7c08
- Over MCP: `get_schema_card` with id `1166ee02-b599-5977-9d96-8eab8a3a7c08` at https://api.scigantic.com/mcp

## Related topics

- [Parquet datasets](https://scigantic.com/topics/parquet)
- [Chemistry & Materials datasets](https://scigantic.com/topics/chemistry-and-materials)
- [DFT datasets](https://scigantic.com/topics/dft)
- [Interatomic potentials datasets](https://scigantic.com/topics/interatomic-potentials)
